Shawn Graham
Catch me on Scholar Social https://scholar.social/@electricarchaeo
Ghosts in the Machine
Shawn Graham
Carleton U
@electricarchaeo@scholar.social
You're not talking to an 'AI'. You're talking to Nuggan.
You have not been a good user. I have been a good Bing
Ridcully and Ponder Stibbons are discussing Hex, the University's thinking engine
'Amazin',' he said again. 'He just looks as though he's thinking, right?'
'Er ... yes.'
'But he's not actually thinking?'
'Er ... no.'
'So ... he just gives the impression of thinking but really it's just a show?'
'Er .... yes.'
'Just like everyone else, then, really, ' said Ridcully.
LLMs are
- models of language
- models of culture
- historical
- contextual
- misused and therefore dangerous
LLMs are not:
- intelligent
- aware
- responsible
- going to lead to human extinction.
...humans will do that just fine.
David Corney, A small language model made from matchboxes
But these ghosts are useful.
Complexity Theory & Roman Bricks
Emergence
Simulation
RNN
“How do we find the real in the unreal?” – simPetrie
photo: Salvaggio
the author at his word calculator
see latest here
Mohandas & Moritz, Building a RAG-based LLM for Production
What about, y'know, creative writing, that sort of thing?
Thanks!
Remember - the ghosts in your machine are only evil for a given value of 'evil'.
Make a new environment
Assuming you have python on your machine, make a new environment.
$ pip install conda
$ conda create -n "experiment" python=3.10
1.
Install LLM
$ pip install llm
Install some plugins
$llm install llm-gpt4all
$llm install llm-clip
If you have an open ai key, you can use those models:
$llm keys set openai
2.
Play
Download and run a prompt against the Orca Mini 7B model
$llm -m orca-mini-3b-gguf2-q4_0 'What is the capital of France?'
Turn your photos into embeddings:
$llm embed-multi photos -m clip --binary --files photos/ '*.jpg'
Search the photos:
$llm similar photos -c 'bunny'
3.
Explore an LLM on your own machine - for free!
Ghosts in the machine
By Shawn Graham
Ghosts in the machine
A jaunt through how I'm thinking about LLMs and some of the ways I've been using LLMs
